From a5f7a5bb3b7f28ba7e4fa246775b29a0e5537255 Mon Sep 17 00:00:00 2001 From: Jinke Han Date: Tue, 8 Sep 2026 15:37:42 +0800 Subject: x86/kprobes: Fix crash when probing CS CALL instructions When using eBPF to probe CS CALL instructions within a function, a crash can be triggered. The eBPF tool probes offset 257 of the __hrtimer_run_queues() function: <__hrtimer_run_queues+249>: nopl 0x0(%rax,%rax,1) <__hrtimer_run_queues+254>: mov %r14,%rdi <__hrtimer_run_queues+257>: cs call <__x86_indirect_thunk_r12> <__hrtimer_run_queues+263>: mov %eax,%r12d <__hrtimer_run_queues+266>: xchg %ax,%ax <__hrtimer_run_queues+268>: mov %r13,%rdi Which triggers this crash: BUG: unable to handle page fault for address: 00000000000f41c9 #PF: supervisor write access in kernel mode #PF: error_code(0x0002) - not-present page PGD 0 P4D 0 Oops: 0002 [#1] SMP NOPTI CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P RIP: 0010:__hrtimer_run_queues+0x106/0x230 Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is at the 6th byte of the above CS CALL instruction. Since the CS CALL instruction occupies 6 bytes, the exception occurred in the middle of that call instruction. The root cause is that when using eBPF tools to probe in the middle of a function, a kprobe with INT3 is used as the underlying implementation. During single-step emulation of the original CALL instruction, int3_emulate_call() assumes that the probed CALL instruction is 5 bytes long. However, the actual CS-prefixed CALL instruction occupies 6 bytes, so it constructs an incorrect exception return address. When the CPU returns from the kprobe handler, the next instruction to be executed is at the address of the last byte of that CS CALL instruction. Coincidentally, starting from that address, the CPU fetches and decodes a completely different instruction, which ultimately triggers a kernel crash. Fix the issue by using the actual instruction length obtained from the instruction decoder when constructing the exception return address, rather than relying on the hardcoded CALL_INSN_SIZE macro. [ mingo: Refined the changelog ] Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step") Suggested-by: Masami Hiramatsu (Google) Signed-off-by: Jinke Han Signed-off-by: Ingo Molnar Reviewed-by: Masami Hiramatsu (Google) Acked-by: Yafang Shao Acked-by: Borislav Petkov Cc: Peter Zijlstra Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000 --- net/sched/sch_plug.c | 230 +++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 230 insertions(+) create mode 100644 net/sched/sch_plug.c (limited to 'net/sched/sch_plug.c') diff --git a/net/sched/sch_plug.c b/net/sched/sch_plug.c new file mode 100644 index 000000000..b60ddfee6 --- /dev/null +++ b/net/sched/sch_plug.c @@ -0,0 +1,230 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * sch_plug.c Queue traffic until an explicit release command + * + * There are two ways to use this qdisc: + * 1. A simple "instantaneous" plug/unplug operation, by issuing an alternating + * sequence of TCQ_PLUG_BUFFER & TCQ_PLUG_RELEASE_INDEFINITE commands. + * + * 2. For network output buffering (a.k.a output commit) functionality. + * Output commit property is commonly used by applications using checkpoint + * based fault-tolerance to ensure that the checkpoint from which a system + * is being restored is consistent w.r.t outside world. + * + * Consider for e.g. Remus - a Virtual Machine checkpointing system, + * wherein a VM is checkpointed, say every 50ms. The checkpoint is replicated + * asynchronously to the backup host, while the VM continues executing the + * next epoch speculatively. + * + * The following is a typical sequence of output buffer operations: + * 1.At epoch i, start_buffer(i) + * 2. At end of epoch i (i.e. after 50ms): + * 2.1 Stop VM and take checkpoint(i). + * 2.2 start_buffer(i+1) and Resume VM + * 3. While speculatively executing epoch(i+1), asynchronously replicate + * checkpoint(i) to backup host. + * 4. When checkpoint_ack(i) is received from backup, release_buffer(i) + * Thus, this Qdisc would receive the following sequence of commands: + * TCQ_PLUG_BUFFER (epoch i) + * .. TCQ_PLUG_BUFFER (epoch i+1) + * ....TCQ_PLUG_RELEASE_ONE (epoch i) + * ......TCQ_PLUG_BUFFER (epoch i+2) + * ........ + */ + +#include +#include +#include +#include +#include +#include +#include + +/* + * State of the queue, when used for network output buffering: + * + * plug(i+1) plug(i) head + * ------------------+--------------------+----------------> + * | | + * | | + * pkts_current_epoch| pkts_last_epoch |pkts_to_release + * ----------------->|<--------+--------->|+---------------> + * v v + * + */ + +struct plug_sched_data { + /* If true, the dequeue function releases all packets + * from head to end of the queue. The queue turns into + * a pass-through queue for newly arriving packets. + */ + bool unplug_indefinite; + + bool throttled; + + /* Queue Limit in bytes */ + u32 limit; + + /* Number of packets (output) from the current speculatively + * executing epoch. + */ + u32 pkts_current_epoch; + + /* Number of packets corresponding to the recently finished + * epoch. These will be released when we receive a + * TCQ_PLUG_RELEASE_ONE command. This command is typically + * issued after committing a checkpoint at the target. + */ + u32 pkts_last_epoch; + + /* + * Number of packets from the head of the queue, that can + * be released (committed checkpoint). + */ + u32 pkts_to_release; +}; + +static int plug_enqueue(struct sk_buff *skb, struct Qdisc *sch, + struct sk_buff **to_free) +{ + struct plug_sched_data *q = qdisc_priv(sch); + + if (likely((u64)sch->qstats.backlog + skb->len <= q->limit)) { + if (!q->unplug_indefinite) + q->pkts_current_epoch++; + return qdisc_enqueue_tail(skb, sch); + } + + return qdisc_drop(skb, sch, to_free); +} + +static struct sk_buff *plug_dequeue(struct Qdisc *sch) +{ + struct plug_sched_data *q = qdisc_priv(sch); + + if (q->throttled) + return NULL; + + if (!q->unplug_indefinite) { + if (!q->pkts_to_release) { + /* No more packets to dequeue. Block the queue + * and wait for the next release command. + */ + q->throttled = true; + return NULL; + } + q->pkts_to_release--; + } + + return qdisc_dequeue_head(sch); +} + +static int plug_init(struct Qdisc *sch, struct nlattr *opt, + struct netlink_ext_ack *extack) +{ + struct plug_sched_data *q = qdisc_priv(sch); + + q->pkts_current_epoch = 0; + q->pkts_last_epoch = 0; + q->pkts_to_release = 0; + q->unplug_indefinite = false; + + if (opt == NULL) { + q->limit = qdisc_dev(sch)->tx_queue_len + * psched_mtu(qdisc_dev(sch)); + } else { + struct tc_plug_qopt *ctl = nla_data(opt); + + if (nla_len(opt) < sizeof(*ctl)) + return -EINVAL; + + q->limit = ctl->limit; + } + + q->throttled = true; + return 0; +} + +/* Receives 4 types of messages: + * TCQ_PLUG_BUFFER: Inset a plug into the queue and + * buffer any incoming packets + * TCQ_PLUG_RELEASE_ONE: Dequeue packets from queue head + * to beginning of the next plug. + * TCQ_PLUG_RELEASE_INDEFINITE: Dequeue all packets from queue. + * Stop buffering packets until the next TCQ_PLUG_BUFFER + * command is received (just act as a pass-thru queue). + * TCQ_PLUG_LIMIT: Increase/decrease queue size + */ +static int plug_change(struct Qdisc *sch, struct nlattr *opt, + struct netlink_ext_ack *extack) +{ + struct plug_sched_data *q = qdisc_priv(sch); + struct tc_plug_qopt *msg; + + msg = nla_data(opt); + if (nla_len(opt) < sizeof(*msg)) + return -EINVAL; + + switch (msg->action) { + case TCQ_PLUG_BUFFER: + /* Save size of the current buffer */ + q->pkts_last_epoch = q->pkts_current_epoch; + q->pkts_current_epoch = 0; + if (q->unplug_indefinite) + q->throttled = true; + q->unplug_indefinite = false; + break; + case TCQ_PLUG_RELEASE_ONE: + /* Add packets from the last complete buffer to the + * packets to be released set. + */ + q->pkts_to_release += q->pkts_last_epoch; + q->pkts_last_epoch = 0; + q->throttled = false; + netif_schedule_queue(sch->dev_queue); + break; + case TCQ_PLUG_RELEASE_INDEFINITE: + q->unplug_indefinite = true; + q->pkts_to_release = 0; + q->pkts_last_epoch = 0; + q->pkts_current_epoch = 0; + q->throttled = false; + netif_schedule_queue(sch->dev_queue); + break; + case TCQ_PLUG_LIMIT: + /* Limit is supplied in bytes */ + q->limit = msg->limit; + break; + default: + return -EINVAL; + } + + return 0; +} + +static struct Qdisc_ops plug_qdisc_ops __read_mostly = { + .id = "plug", + .priv_size = sizeof(struct plug_sched_data), + .enqueue = plug_enqueue, + .dequeue = plug_dequeue, + .peek = qdisc_peek_dequeued, + .init = plug_init, + .change = plug_change, + .reset = qdisc_reset_queue, + .owner = THIS_MODULE, +}; +MODULE_ALIAS_NET_SCH("plug"); + +static int __init plug_module_init(void) +{ + return register_qdisc(&plug_qdisc_ops); +} + +static void __exit plug_module_exit(void) +{ + unregister_qdisc(&plug_qdisc_ops); +} +module_init(plug_module_init) +module_exit(plug_module_exit) +MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION("Qdisc to plug and unplug traffic via netlink control"); -- cgit v1.3.1